Dear colleagues, Currently, we install fuel-libraryX.Y package(s) on the master node and then right before starting actual deployment we rsync [1] puppet modules (one of installed versions) from the master node to slave nodes. Such a flow makes things much more complicated than they could be if we installed puppet modules on slave nodes as rpm/deb packages. Deployment itself is parameterized by repo urls (upstream + mos) and this pre-deployment task could be nothing more than just installing fuel-library package from mos repo defined for a cluster. We would not have several versions of fuel-library on the master node, we would not need that complicated upgrade stuff like we currently have for puppet modules. Please give your opinions on this. [1] https://github.com/stackforge/fuel-web/blob/master/nailgun/nailgun/orchestrator/tasks_serializer.py#L205-L218 Vladimir Kozhukalov -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20150908/d715caea/attachment.html>